Attacks on GenAI Data and Using Vector Encryption for Protection

Learn about critical security vulnerabilities in GenAI systems and innovative solutions through this DEF CON 32 conference presentation. Explore the emerging challenges as vector databases increasingly store sensitive data like health records, financial information, and HR data without adequate security measures. Discover how embedding inversion attacks can reverse engineer vectors to reconstruct faces, sentences, and images. Delve into vector encryption, a groundbreaking cryptographic approach that enables privacy-preserving searches on encrypted vectors. Examine the advantages, limitations, and current developments in this field, including open-source implementations designed to address these security concerns.